Re: Radon - county linear model and LNT

[BERNARD L COHEN <blc+@PITT.EDU>]

On Fri, 27 Jun 2003, Otto G. Raabe wrote:



> I think that Lubin is saying that when you plot lung cancer rates versus

> radon level you are indirectly assuming a linear relationships between the

> overall lung cancer rate and all of the co-variates that may affect the

> lung cancer rate. Smoking is an obvious co-variate, but there may be others

> such as genetic susceptibility, dietary level of certain nutrients, etc.



	--I treat co-variates (which I call confounding factors, CF) by

stratification. That is, I do a complete analysis for all counties with

similar values of the CF and get a result. I do this several groups of

counties with different similar values of the CF. In practice, I do this

dy dividing the counties into 10 deciles, the 10% with the largest values

of the CF, the 10% with the next largest values of the CF, etc ending with

the 10% with the smallest value of the CF. Each of these 10 deciles is a

data subset, and I do a complete analysis for each decile. I then average

the results for the 10 deciles, and observe if there is a significant

trend in the results as one goes from the first to the last decile.

FOR EACH DECILE, THE VALUES OF THE CF ARE ABOUT THE SAME, SO THEY CAN

CAUSE VERY LITTLE CONFOUNDING.

	This avoids any assumption about how the results depend on the CF,

linear or otherwise. This is explained and justified in Sec. 2.1 of item

#7 on my web site, and elaborated in the following sections. I have

studied over 500 possible confounding factors, all that I could get values

for, but anything else I can think of correlates strongly with some of

these. Can you suggest specific ones that should be considered and are

considered in case-control studies?

	Note that epidemiologists commonly use multivariate regression in

their case-control studies, and this does assume linear dependences.



************************************************************************

You are currently subscribed to the Radsafe mailing list. To unsubscribe,

send an e-mail to Majordomo@list.vanderbilt.edu  Put the text "unsubscribe

radsafe" (no quote marks) in the body of the e-mail, with no subject line.

You can view the Radsafe archives at http://www.vanderbilt.edu/radsafe/